diff --git a/src/nvtt/squish/weightedclusterfit.cpp b/src/nvtt/squish/weightedclusterfit.cpp
index 36186b8..018a3ef 100644
--- a/src/nvtt/squish/weightedclusterfit.cpp
+++ b/src/nvtt/squish/weightedclusterfit.cpp
@@ -135,6 +135,8 @@ void WeightedClusterFit::Compress3( void* block )
 	Vec4 const zero = VEC4_CONST(0.0f);
 	Vec4 const half(0.5f, 0.5f, 0.5f, 0.25f);
 	Vec4 const two = VEC4_CONST(2.0);
+	Vec4 const grid( 31.0f, 63.0f, 31.0f, 0.0f );
+	Vec4 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
 	 
 	// declare variables
 	Vec4 beststart = VEC4_CONST( 0.0f );
@@ -173,24 +175,21 @@ void WeightedClusterFit::Compress3( void* block )
 			Vec4 a = NegativeMultiplySubtract(betax_sum, alphabeta_sum, alphax_sum*beta2_sum) * factor;
 			Vec4 b = NegativeMultiplySubtract(alphax_sum, alphabeta_sum, betax_sum*alpha2_sum) * factor;
 			
-			// clamp the output to [0, 1]
+			// clamp to the grid
 			a = Min( one, Max( zero, a ) );
 			b = Min( one, Max( zero, b ) );
-			
-			// clamp to the grid
-			Vec4 const grid( 31.0f, 63.0f, 31.0f, 0.0f );
-			Vec4 const gridrcp( 0.03227752766457f, 0.01583151765563f, 0.03227752766457f, 0.0f );
 			a = Truncate( MultiplyAdd( grid, a, half ) ) * gridrcp;
 			b = Truncate( MultiplyAdd( grid, b, half ) ) * gridrcp;
 			
-			// compute the error
-			Vec4 e1 = MultiplyAdd( a, alphax_sum, b*betax_sum );
-			Vec4 e2 = MultiplyAdd( a*a, alpha2_sum, b*b*beta2_sum );
-			Vec4 e3 = MultiplyAdd( a*b*alphabeta_sum - e1, two, e2 );
-			
+			// compute the error (we skip the constant xxsum)
+			Vec4 e1 = MultiplyAdd( a*a, alpha2_sum, b*b*beta2_sum );
+			Vec4 e2 = NegativeMultiplySubtract( a, alphax_sum, a*b*alphabeta_sum );
+			Vec4 e3 = NegativeMultiplySubtract( b, betax_sum, e2 );
+			Vec4 e4 = MultiplyAdd( two, e3, e1 );
+
 			// apply the metric to the error term
-			Vec4 e4 = e3 * m_metricSqr;
-			Vec4 error = e4.SplatX() + e4.SplatY() + e4.SplatZ();
+			Vec4 e5 = e4 * m_metricSqr;
+			Vec4 error = e5.SplatX() + e5.SplatY() + e5.SplatZ();
 			
 			// keep the solution if it wins
 			if( CompareAnyLessThan( error, besterror ) )
@@ -250,6 +249,9 @@ void WeightedClusterFit::Compress4( void* block )
 	Vec4 const two = VEC4_CONST(2.0);
 	Vec4 const onethird( 1.0f/3.0f, 1.0f/3.0f, 1.0f/3.0f, 1.0f/9.0f );
 	Vec4 const twothirds( 2.0f/3.0f, 2.0f/3.0f, 2.0f/3.0f, 4.0f/9.0f );
+    Vec4 const twonineths = VEC4_CONST( 2.0f/9.0f );
+	Vec4 const grid( 31.0f, 63.0f, 31.0f, 0.0f );
+	Vec4 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f, 0.0f );
 	
 	// declare variables
 	Vec4 beststart = VEC4_CONST( 0.0f );
@@ -274,16 +276,16 @@ void WeightedClusterFit::Compress4( void* block )
 				
 				//Vec3 const alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
 				//float const alpha2_sum = w0 + w1 * (4.0f/9.0f) + w2 * (1.0f/9.0f);
-				Vec4 const alphax_sum = x0 + MultiplyAdd(x1, twothirds, x2 * onethird); // alphax_sum, alpha2_sum
+                Vec4 const alphax_sum = MultiplyAdd(x2, onethird, MultiplyAdd(x1, twothirds, x0)); // alphax_sum, alpha2_sum
 				Vec4 const alpha2_sum = alphax_sum.SplatW();
 				
 				//Vec3 const betax_sum = x3 + x2 * (2.0f / 3.0f) + x1 * (1.0f / 3.0f);
 				//float const beta2_sum = w3 + w2 * (4.0f/9.0f) + w1 * (1.0f/9.0f);
-				Vec4 const betax_sum = x3 + MultiplyAdd(x2, twothirds, x1 * onethird); // betax_sum, beta2_sum
+				Vec4 const betax_sum = MultiplyAdd(x2, twothirds, MultiplyAdd(x1, onethird, x3)); // betax_sum, beta2_sum
 				Vec4 const beta2_sum = betax_sum.SplatW();
 				
-				//float const alphabeta_sum = w1 * (2.0f/9.0f) + w2 * (2.0f/9.0f);
-				Vec4 const alphabeta_sum = two * (x1 * onethird + x2 * onethird).SplatW(); // alphabeta_sum
+				//float const alphabeta_sum = (w1 + w2) * (2.0f/9.0f);
+                Vec4 const alphabeta_sum = twonineths*( x1 + x2 ).SplatW(); // alphabeta_sum
 				
 				// float const factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
 				Vec4 const factor = Reciprocal( NegativeMultiplySubtract(alphabeta_sum, alphabeta_sum, alpha2_sum*beta2_sum) );
@@ -291,24 +293,21 @@ void WeightedClusterFit::Compress4( void* block )
 				Vec4 a = NegativeMultiplySubtract(betax_sum, alphabeta_sum, alphax_sum*beta2_sum) * factor;
 				Vec4 b = NegativeMultiplySubtract(alphax_sum, alphabeta_sum, betax_sum*alpha2_sum) * factor;
 				
-				// clamp the output to [0, 1]
+				// clamp to the grid
 				a = Min( one, Max( zero, a ) );
 				b = Min( one, Max( zero, b ) );
-				
-				// clamp to the grid
-				Vec4 const grid( 31.0f, 63.0f, 31.0f, 0.0f );
-				Vec4 const gridrcp( 0.03227752766457f, 0.01583151765563f, 0.03227752766457f, 0.0f );
 				a = Truncate( MultiplyAdd( grid, a, half ) ) * gridrcp;
 				b = Truncate( MultiplyAdd( grid, b, half ) ) * gridrcp;
 				
-				// compute the error
-				Vec4 e1 = MultiplyAdd( a, alphax_sum, b*betax_sum );
-				Vec4 e2 = MultiplyAdd( a*a, alpha2_sum, b*b*beta2_sum );
-				Vec4 e3 = MultiplyAdd( a*b*alphabeta_sum - e1, two, e2 );
-				
+				// compute the error (we skip the constant xxsum)
+				Vec4 e1 = MultiplyAdd( a*a, alpha2_sum, b*b*beta2_sum );
+				Vec4 e2 = NegativeMultiplySubtract( a, alphax_sum, a*b*alphabeta_sum );
+				Vec4 e3 = NegativeMultiplySubtract( b, betax_sum, e2 );
+				Vec4 e4 = MultiplyAdd( two, e3, e1 );
+
 				// apply the metric to the error term
-				Vec4 e4 = e3 * m_metricSqr;
-				Vec4 error = e4.SplatX() + e4.SplatY() + e4.SplatZ();
+				Vec4 e5 = e4 * m_metricSqr;
+				Vec4 error = e5.SplatX() + e5.SplatY() + e5.SplatZ();
 				
 				// keep the solution if it wins
 				if( CompareAnyLessThan( error, besterror ) )
@@ -368,6 +367,12 @@ void WeightedClusterFit::Compress4( void* block )
 
 void WeightedClusterFit::Compress3( void* block )
 {
+	Vec3 const one( 1.0f );
+	Vec3 const zero( 0.0f );
+	Vec3 const half( 0.5f );
+    Vec3 const grid( 31.0f, 63.0f, 31.0f );
+    Vec3 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f );
+
 	// declare variables
 	Vec3 beststart( 0.0f );
 	Vec3 bestend( 0.0f );
@@ -400,16 +405,9 @@ void WeightedClusterFit::Compress3( void* block )
 			Vec3 a = (alphax_sum*beta2_sum - betax_sum*alphabeta_sum) * factor;
 			Vec3 b = (betax_sum*alpha2_sum - alphax_sum*alphabeta_sum) * factor;
 			
-			// clamp the output to [0, 1]
-			Vec3 const one( 1.0f );
-			Vec3 const zero( 0.0f );
+			// clamp to the grid
 			a = Min( one, Max( zero, a ) );
 			b = Min( one, Max( zero, b ) );
-			
-			// clamp to the grid
-			Vec3 const grid( 31.0f, 63.0f, 31.0f );
-			Vec3 const gridrcp( 0.03227752766457f, 0.01583151765563f, 0.03227752766457f );
-			Vec3 const half( 0.5f );
 			a = Floor( grid*a + half )*gridrcp;
 			b = Floor( grid*b + half )*gridrcp;
 			
@@ -470,6 +468,12 @@ void WeightedClusterFit::Compress3( void* block )
 
 void WeightedClusterFit::Compress4( void* block )
 {
+	Vec3 const one( 1.0f );
+	Vec3 const zero( 0.0f );
+	Vec3 const half( 0.5f );
+	Vec3 const grid( 31.0f, 63.0f, 31.0f );
+	Vec3 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f );
+
 	// declare variables
 	Vec3 beststart( 0.0f );
 	Vec3 bestend( 0.0f );
@@ -505,16 +509,9 @@ void WeightedClusterFit::Compress4( void* block )
 				Vec3 a = ( alphax_sum*beta2_sum - betax_sum*alphabeta_sum )*factor;
 				Vec3 b = ( betax_sum*alpha2_sum - alphax_sum*alphabeta_sum )*factor;
 				
-				// clamp the output to [0, 1]
-				Vec3 const one( 1.0f );
-				Vec3 const zero( 0.0f );
+				// clamp to the grid
 				a = Min( one, Max( zero, a ) );
 				b = Min( one, Max( zero, b ) );
-				
-				// clamp to the grid
-				Vec3 const grid( 31.0f, 63.0f, 31.0f );
-				Vec3 const gridrcp( 0.03227752766457f, 0.01583151765563f, 0.03227752766457f );
-				Vec3 const half( 0.5f );
 				a = Floor( grid*a + half )*gridrcp;
 				b = Floor( grid*b + half )*gridrcp;